xref: /freebsd/sys/fs/nfs/nfs_commonkrpc.c (revision 389e4940069316fe667ffa263fa7d6390d0a960f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1989, 1991, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Rick Macklem at The University of Guelph.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 /*
40  * Socket operations for use by nfs
41  */
42 
43 #include "opt_kgssapi.h"
44 #include "opt_nfs.h"
45 
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/kernel.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mbuf.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/proc.h>
56 #include <sys/signalvar.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysctl.h>
59 #include <sys/syslog.h>
60 #include <sys/vnode.h>
61 
62 #include <rpc/rpc.h>
63 #include <rpc/krpc.h>
64 
65 #include <kgssapi/krb5/kcrypto.h>
66 
67 #include <fs/nfs/nfsport.h>
68 
69 #ifdef KDTRACE_HOOKS
70 #include <sys/dtrace_bsd.h>
71 
72 dtrace_nfsclient_nfs23_start_probe_func_t
73 		dtrace_nfscl_nfs234_start_probe;
74 
75 dtrace_nfsclient_nfs23_done_probe_func_t
76 		dtrace_nfscl_nfs234_done_probe;
77 
78 /*
79  * Registered probes by RPC type.
80  */
81 uint32_t	nfscl_nfs2_start_probes[NFSV41_NPROCS + 1];
82 uint32_t	nfscl_nfs2_done_probes[NFSV41_NPROCS + 1];
83 
84 uint32_t	nfscl_nfs3_start_probes[NFSV41_NPROCS + 1];
85 uint32_t	nfscl_nfs3_done_probes[NFSV41_NPROCS + 1];
86 
87 uint32_t	nfscl_nfs4_start_probes[NFSV41_NPROCS + 1];
88 uint32_t	nfscl_nfs4_done_probes[NFSV41_NPROCS + 1];
89 #endif
90 
91 NFSSTATESPINLOCK;
92 NFSREQSPINLOCK;
93 NFSDLOCKMUTEX;
94 NFSCLSTATEMUTEX;
95 extern struct nfsstatsv1 nfsstatsv1;
96 extern struct nfsreqhead nfsd_reqq;
97 extern int nfscl_ticks;
98 extern void (*ncl_call_invalcaches)(struct vnode *);
99 extern int nfs_numnfscbd;
100 extern int nfscl_debuglevel;
101 
102 SVCPOOL		*nfscbd_pool;
103 static int	nfsrv_gsscallbackson = 0;
104 static int	nfs_bufpackets = 4;
105 static int	nfs_reconnects;
106 static int	nfs3_jukebox_delay = 10;
107 static int	nfs_skip_wcc_data_onerr = 1;
108 
109 SYSCTL_DECL(_vfs_nfs);
110 
111 SYSCTL_INT(_vfs_nfs, OID_AUTO, bufpackets, CTLFLAG_RW, &nfs_bufpackets, 0,
112     "Buffer reservation size 2 < x < 64");
113 SYSCTL_INT(_vfs_nfs, OID_AUTO, reconnects, CTLFLAG_RD, &nfs_reconnects, 0,
114     "Number of times the nfs client has had to reconnect");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs3_jukebox_delay, CTLFLAG_RW, &nfs3_jukebox_delay, 0,
116     "Number of seconds to delay a retry after receiving EJUKEBOX");
117 SYSCTL_INT(_vfs_nfs, OID_AUTO, skip_wcc_data_onerr, CTLFLAG_RW, &nfs_skip_wcc_data_onerr, 0,
118     "Disable weak cache consistency checking when server returns an error");
119 
120 static void	nfs_down(struct nfsmount *, struct thread *, const char *,
121     int, int);
122 static void	nfs_up(struct nfsmount *, struct thread *, const char *,
123     int, int);
124 static int	nfs_msg(struct thread *, const char *, const char *, int);
125 
126 struct nfs_cached_auth {
127 	int		ca_refs; /* refcount, including 1 from the cache */
128 	uid_t		ca_uid;	 /* uid that corresponds to this auth */
129 	AUTH		*ca_auth; /* RPC auth handle */
130 };
131 
132 static int nfsv2_procid[NFS_V3NPROCS] = {
133 	NFSV2PROC_NULL,
134 	NFSV2PROC_GETATTR,
135 	NFSV2PROC_SETATTR,
136 	NFSV2PROC_LOOKUP,
137 	NFSV2PROC_NOOP,
138 	NFSV2PROC_READLINK,
139 	NFSV2PROC_READ,
140 	NFSV2PROC_WRITE,
141 	NFSV2PROC_CREATE,
142 	NFSV2PROC_MKDIR,
143 	NFSV2PROC_SYMLINK,
144 	NFSV2PROC_CREATE,
145 	NFSV2PROC_REMOVE,
146 	NFSV2PROC_RMDIR,
147 	NFSV2PROC_RENAME,
148 	NFSV2PROC_LINK,
149 	NFSV2PROC_READDIR,
150 	NFSV2PROC_NOOP,
151 	NFSV2PROC_STATFS,
152 	NFSV2PROC_NOOP,
153 	NFSV2PROC_NOOP,
154 	NFSV2PROC_NOOP,
155 };
156 
157 /*
158  * Initialize sockets and congestion for a new NFS connection.
159  * We do not free the sockaddr if error.
160  * Which arguments are set to NULL indicate what kind of call it is.
161  * cred == NULL --> a call to connect to a pNFS DS
162  * nmp == NULL --> indicates an upcall to userland or a NFSv4 callback
163  */
164 int
165 newnfs_connect(struct nfsmount *nmp, struct nfssockreq *nrp,
166     struct ucred *cred, NFSPROC_T *p, int callback_retry_mult)
167 {
168 	int rcvreserve, sndreserve;
169 	int pktscale, pktscalesav;
170 	struct sockaddr *saddr;
171 	struct ucred *origcred;
172 	CLIENT *client;
173 	struct netconfig *nconf;
174 	struct socket *so;
175 	int one = 1, retries, error = 0;
176 	struct thread *td = curthread;
177 	SVCXPRT *xprt;
178 	struct timeval timo;
179 
180 	/*
181 	 * We need to establish the socket using the credentials of
182 	 * the mountpoint.  Some parts of this process (such as
183 	 * sobind() and soconnect()) will use the curent thread's
184 	 * credential instead of the socket credential.  To work
185 	 * around this, temporarily change the current thread's
186 	 * credential to that of the mountpoint.
187 	 *
188 	 * XXX: It would be better to explicitly pass the correct
189 	 * credential to sobind() and soconnect().
190 	 */
191 	origcred = td->td_ucred;
192 
193 	/*
194 	 * Use the credential in nr_cred, if not NULL.
195 	 */
196 	if (nrp->nr_cred != NULL)
197 		td->td_ucred = nrp->nr_cred;
198 	else
199 		td->td_ucred = cred;
200 	saddr = nrp->nr_nam;
201 
202 	if (saddr->sa_family == AF_INET)
203 		if (nrp->nr_sotype == SOCK_DGRAM)
204 			nconf = getnetconfigent("udp");
205 		else
206 			nconf = getnetconfigent("tcp");
207 	else if (saddr->sa_family == AF_LOCAL)
208 		nconf = getnetconfigent("local");
209 	else
210 		if (nrp->nr_sotype == SOCK_DGRAM)
211 			nconf = getnetconfigent("udp6");
212 		else
213 			nconf = getnetconfigent("tcp6");
214 
215 	pktscale = nfs_bufpackets;
216 	if (pktscale < 2)
217 		pktscale = 2;
218 	if (pktscale > 64)
219 		pktscale = 64;
220 	pktscalesav = pktscale;
221 	/*
222 	 * soreserve() can fail if sb_max is too small, so shrink pktscale
223 	 * and try again if there is an error.
224 	 * Print a log message suggesting increasing sb_max.
225 	 * Creating a socket and doing this is necessary since, if the
226 	 * reservation sizes are too large and will make soreserve() fail,
227 	 * the connection will work until a large send is attempted and
228 	 * then it will loop in the krpc code.
229 	 */
230 	so = NULL;
231 	saddr = NFSSOCKADDR(nrp->nr_nam, struct sockaddr *);
232 	error = socreate(saddr->sa_family, &so, nrp->nr_sotype,
233 	    nrp->nr_soproto, td->td_ucred, td);
234 	if (error) {
235 		td->td_ucred = origcred;
236 		goto out;
237 	}
238 	do {
239 	    if (error != 0 && pktscale > 2) {
240 		if (nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
241 		    pktscale == pktscalesav)
242 		    printf("Consider increasing kern.ipc.maxsockbuf\n");
243 		pktscale--;
244 	    }
245 	    if (nrp->nr_sotype == SOCK_DGRAM) {
246 		if (nmp != NULL) {
247 			sndreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
248 			    pktscale;
249 			rcvreserve = (NFS_MAXDGRAMDATA + NFS_MAXPKTHDR) *
250 			    pktscale;
251 		} else {
252 			sndreserve = rcvreserve = 1024 * pktscale;
253 		}
254 	    } else {
255 		if (nrp->nr_sotype != SOCK_STREAM)
256 			panic("nfscon sotype");
257 		if (nmp != NULL) {
258 			sndreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
259 			    sizeof (u_int32_t)) * pktscale;
260 			rcvreserve = (NFS_MAXBSIZE + NFS_MAXXDR +
261 			    sizeof (u_int32_t)) * pktscale;
262 		} else {
263 			sndreserve = rcvreserve = 1024 * pktscale;
264 		}
265 	    }
266 	    error = soreserve(so, sndreserve, rcvreserve);
267 	    if (error != 0 && nmp != NULL && nrp->nr_sotype == SOCK_STREAM &&
268 		pktscale <= 2)
269 		printf("Must increase kern.ipc.maxsockbuf or reduce"
270 		    " rsize, wsize\n");
271 	} while (error != 0 && pktscale > 2);
272 	soclose(so);
273 	if (error) {
274 		td->td_ucred = origcred;
275 		goto out;
276 	}
277 
278 	client = clnt_reconnect_create(nconf, saddr, nrp->nr_prog,
279 	    nrp->nr_vers, sndreserve, rcvreserve);
280 	CLNT_CONTROL(client, CLSET_WAITCHAN, "nfsreq");
281 	if (nmp != NULL) {
282 		if ((nmp->nm_flag & NFSMNT_INT))
283 			CLNT_CONTROL(client, CLSET_INTERRUPTIBLE, &one);
284 		if ((nmp->nm_flag & NFSMNT_RESVPORT))
285 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
286 		if (NFSHASSOFT(nmp)) {
287 			if (nmp->nm_sotype == SOCK_DGRAM)
288 				/*
289 				 * For UDP, the large timeout for a reconnect
290 				 * will be set to "nm_retry * nm_timeo / 2", so
291 				 * we only want to do 2 reconnect timeout
292 				 * retries.
293 				 */
294 				retries = 2;
295 			else
296 				retries = nmp->nm_retry;
297 		} else
298 			retries = INT_MAX;
299 		if (NFSHASNFSV4N(nmp)) {
300 			if (cred != NULL) {
301 				/*
302 				 * Make sure the nfscbd_pool doesn't get
303 				 * destroyed while doing this.
304 				 */
305 				NFSD_LOCK();
306 				if (nfs_numnfscbd > 0) {
307 					nfs_numnfscbd++;
308 					NFSD_UNLOCK();
309 					xprt = svc_vc_create_backchannel(
310 					    nfscbd_pool);
311 					CLNT_CONTROL(client, CLSET_BACKCHANNEL,
312 					    xprt);
313 					NFSD_LOCK();
314 					nfs_numnfscbd--;
315 					if (nfs_numnfscbd == 0)
316 						wakeup(&nfs_numnfscbd);
317 				}
318 				NFSD_UNLOCK();
319 			} else {
320 				/*
321 				 * cred == NULL for a DS connect.
322 				 * For connects to a DS, set a retry limit
323 				 * so that failed DSs will be detected.
324 				 * This is ok for NFSv4.1, since a DS does
325 				 * not maintain open/lock state and is the
326 				 * only case where using a "soft" mount is
327 				 * recommended for NFSv4.
328 				 */
329 				retries = 2;
330 			}
331 		}
332 	} else {
333 		/*
334 		 * Three cases:
335 		 * - Null RPC callback to client
336 		 * - Non-Null RPC callback to client, wait a little longer
337 		 * - upcalls to nfsuserd and gssd (clp == NULL)
338 		 */
339 		if (callback_retry_mult == 0) {
340 			retries = NFSV4_UPCALLRETRY;
341 			CLNT_CONTROL(client, CLSET_PRIVPORT, &one);
342 		} else {
343 			retries = NFSV4_CALLBACKRETRY * callback_retry_mult;
344 		}
345 	}
346 	CLNT_CONTROL(client, CLSET_RETRIES, &retries);
347 
348 	if (nmp != NULL) {
349 		/*
350 		 * For UDP, there are 2 timeouts:
351 		 * - CLSET_RETRY_TIMEOUT sets the initial timeout for the timer
352 		 *   that does a retransmit of an RPC request using the same
353 		 *   socket and xid. This is what you normally want to do,
354 		 *   since NFS servers depend on "same xid" for their
355 		 *   Duplicate Request Cache.
356 		 * - timeout specified in CLNT_CALL_MBUF(), which specifies when
357 		 *   retransmits on the same socket should fail and a fresh
358 		 *   socket created. Each of these timeouts counts as one
359 		 *   CLSET_RETRIES as set above.
360 		 * Set the initial retransmit timeout for UDP. This timeout
361 		 * doesn't exist for TCP and the following call just fails,
362 		 * which is ok.
363 		 */
364 		timo.tv_sec = nmp->nm_timeo / NFS_HZ;
365 		timo.tv_usec = (nmp->nm_timeo % NFS_HZ) * 1000000 / NFS_HZ;
366 		CLNT_CONTROL(client, CLSET_RETRY_TIMEOUT, &timo);
367 	}
368 
369 	mtx_lock(&nrp->nr_mtx);
370 	if (nrp->nr_client != NULL) {
371 		mtx_unlock(&nrp->nr_mtx);
372 		/*
373 		 * Someone else already connected.
374 		 */
375 		CLNT_RELEASE(client);
376 	} else {
377 		nrp->nr_client = client;
378 		/*
379 		 * Protocols that do not require connections may be optionally
380 		 * left unconnected for servers that reply from a port other
381 		 * than NFS_PORT.
382 		 */
383 		if (nmp == NULL || (nmp->nm_flag & NFSMNT_NOCONN) == 0) {
384 			mtx_unlock(&nrp->nr_mtx);
385 			CLNT_CONTROL(client, CLSET_CONNECT, &one);
386 		} else
387 			mtx_unlock(&nrp->nr_mtx);
388 	}
389 
390 
391 	/* Restore current thread's credentials. */
392 	td->td_ucred = origcred;
393 
394 out:
395 	NFSEXITCODE(error);
396 	return (error);
397 }
398 
399 /*
400  * NFS disconnect. Clean up and unlink.
401  */
402 void
403 newnfs_disconnect(struct nfssockreq *nrp)
404 {
405 	CLIENT *client;
406 
407 	mtx_lock(&nrp->nr_mtx);
408 	if (nrp->nr_client != NULL) {
409 		client = nrp->nr_client;
410 		nrp->nr_client = NULL;
411 		mtx_unlock(&nrp->nr_mtx);
412 		rpc_gss_secpurge_call(client);
413 		CLNT_CLOSE(client);
414 		CLNT_RELEASE(client);
415 	} else {
416 		mtx_unlock(&nrp->nr_mtx);
417 	}
418 }
419 
420 static AUTH *
421 nfs_getauth(struct nfssockreq *nrp, int secflavour, char *clnt_principal,
422     char *srv_principal, gss_OID mech_oid, struct ucred *cred)
423 {
424 	rpc_gss_service_t svc;
425 	AUTH *auth;
426 
427 	switch (secflavour) {
428 	case RPCSEC_GSS_KRB5:
429 	case RPCSEC_GSS_KRB5I:
430 	case RPCSEC_GSS_KRB5P:
431 		if (!mech_oid) {
432 			if (!rpc_gss_mech_to_oid_call("kerberosv5", &mech_oid))
433 				return (NULL);
434 		}
435 		if (secflavour == RPCSEC_GSS_KRB5)
436 			svc = rpc_gss_svc_none;
437 		else if (secflavour == RPCSEC_GSS_KRB5I)
438 			svc = rpc_gss_svc_integrity;
439 		else
440 			svc = rpc_gss_svc_privacy;
441 
442 		if (clnt_principal == NULL)
443 			auth = rpc_gss_secfind_call(nrp->nr_client, cred,
444 			    srv_principal, mech_oid, svc);
445 		else {
446 			auth = rpc_gss_seccreate_call(nrp->nr_client, cred,
447 			    clnt_principal, srv_principal, "kerberosv5",
448 			    svc, NULL, NULL, NULL);
449 			return (auth);
450 		}
451 		if (auth != NULL)
452 			return (auth);
453 		/* fallthrough */
454 	case AUTH_SYS:
455 	default:
456 		return (authunix_create(cred));
457 
458 	}
459 }
460 
461 /*
462  * Callback from the RPC code to generate up/down notifications.
463  */
464 
465 struct nfs_feedback_arg {
466 	struct nfsmount *nf_mount;
467 	int		nf_lastmsg;	/* last tprintf */
468 	int		nf_tprintfmsg;
469 	struct thread	*nf_td;
470 };
471 
472 static void
473 nfs_feedback(int type, int proc, void *arg)
474 {
475 	struct nfs_feedback_arg *nf = (struct nfs_feedback_arg *) arg;
476 	struct nfsmount *nmp = nf->nf_mount;
477 	time_t now;
478 
479 	switch (type) {
480 	case FEEDBACK_REXMIT2:
481 	case FEEDBACK_RECONNECT:
482 		now = NFSD_MONOSEC;
483 		if (nf->nf_lastmsg + nmp->nm_tprintf_delay < now) {
484 			nfs_down(nmp, nf->nf_td,
485 			    "not responding", 0, NFSSTA_TIMEO);
486 			nf->nf_tprintfmsg = TRUE;
487 			nf->nf_lastmsg = now;
488 		}
489 		break;
490 
491 	case FEEDBACK_OK:
492 		nfs_up(nf->nf_mount, nf->nf_td,
493 		    "is alive again", NFSSTA_TIMEO, nf->nf_tprintfmsg);
494 		break;
495 	}
496 }
497 
498 /*
499  * newnfs_request - goes something like this
500  *	- does the rpc by calling the krpc layer
501  *	- break down rpc header and return with nfs reply
502  * nb: always frees up nd_mreq mbuf list
503  */
504 int
505 newnfs_request(struct nfsrv_descript *nd, struct nfsmount *nmp,
506     struct nfsclient *clp, struct nfssockreq *nrp, vnode_t vp,
507     struct thread *td, struct ucred *cred, u_int32_t prog, u_int32_t vers,
508     u_char *retsum, int toplevel, u_int64_t *xidp, struct nfsclsession *dssep)
509 {
510 	uint32_t retseq, retval, slotseq, *tl;
511 	time_t waituntil;
512 	int i = 0, j = 0, opcnt, set_sigset = 0, slot;
513 	int error = 0, usegssname = 0, secflavour = AUTH_SYS;
514 	int freeslot, maxslot, reterr, slotpos, timeo;
515 	u_int16_t procnum;
516 	u_int trylater_delay = 1;
517 	struct nfs_feedback_arg nf;
518 	struct timeval timo;
519 	AUTH *auth;
520 	struct rpc_callextra ext;
521 	enum clnt_stat stat;
522 	struct nfsreq *rep = NULL;
523 	char *srv_principal = NULL, *clnt_principal = NULL;
524 	sigset_t oldset;
525 	struct ucred *authcred;
526 	struct nfsclsession *sep;
527 	uint8_t sessionid[NFSX_V4SESSIONID];
528 
529 	sep = dssep;
530 	if (xidp != NULL)
531 		*xidp = 0;
532 	/* Reject requests while attempting a forced unmount. */
533 	if (nmp != NULL && NFSCL_FORCEDISM(nmp->nm_mountp)) {
534 		m_freem(nd->nd_mreq);
535 		return (ESTALE);
536 	}
537 
538 	/*
539 	 * Set authcred, which is used to acquire RPC credentials to
540 	 * the cred argument, by default. The crhold() should not be
541 	 * necessary, but will ensure that some future code change
542 	 * doesn't result in the credential being free'd prematurely.
543 	 */
544 	authcred = crhold(cred);
545 
546 	/* For client side interruptible mounts, mask off the signals. */
547 	if (nmp != NULL && td != NULL && NFSHASINT(nmp)) {
548 		newnfs_set_sigmask(td, &oldset);
549 		set_sigset = 1;
550 	}
551 
552 	/*
553 	 * XXX if not already connected call nfs_connect now. Longer
554 	 * term, change nfs_mount to call nfs_connect unconditionally
555 	 * and let clnt_reconnect_create handle reconnects.
556 	 */
557 	if (nrp->nr_client == NULL)
558 		newnfs_connect(nmp, nrp, cred, td, 0);
559 
560 	/*
561 	 * For a client side mount, nmp is != NULL and clp == NULL. For
562 	 * server calls (callbacks or upcalls), nmp == NULL.
563 	 */
564 	if (clp != NULL) {
565 		NFSLOCKSTATE();
566 		if ((clp->lc_flags & LCL_GSS) && nfsrv_gsscallbackson) {
567 			secflavour = RPCSEC_GSS_KRB5;
568 			if (nd->nd_procnum != NFSPROC_NULL) {
569 				if (clp->lc_flags & LCL_GSSINTEGRITY)
570 					secflavour = RPCSEC_GSS_KRB5I;
571 				else if (clp->lc_flags & LCL_GSSPRIVACY)
572 					secflavour = RPCSEC_GSS_KRB5P;
573 			}
574 		}
575 		NFSUNLOCKSTATE();
576 	} else if (nmp != NULL && NFSHASKERB(nmp) &&
577 	     nd->nd_procnum != NFSPROC_NULL) {
578 		if (NFSHASALLGSSNAME(nmp) && nmp->nm_krbnamelen > 0)
579 			nd->nd_flag |= ND_USEGSSNAME;
580 		if ((nd->nd_flag & ND_USEGSSNAME) != 0) {
581 			/*
582 			 * If there is a client side host based credential,
583 			 * use that, otherwise use the system uid, if set.
584 			 * The system uid is in the nmp->nm_sockreq.nr_cred
585 			 * credentials.
586 			 */
587 			if (nmp->nm_krbnamelen > 0) {
588 				usegssname = 1;
589 				clnt_principal = nmp->nm_krbname;
590 			} else if (nmp->nm_uid != (uid_t)-1) {
591 				KASSERT(nmp->nm_sockreq.nr_cred != NULL,
592 				    ("newnfs_request: NULL nr_cred"));
593 				crfree(authcred);
594 				authcred = crhold(nmp->nm_sockreq.nr_cred);
595 			}
596 		} else if (nmp->nm_krbnamelen == 0 &&
597 		    nmp->nm_uid != (uid_t)-1 && cred->cr_uid == (uid_t)0) {
598 			/*
599 			 * If there is no host based principal name and
600 			 * the system uid is set and this is root, use the
601 			 * system uid, since root won't have user
602 			 * credentials in a credentials cache file.
603 			 * The system uid is in the nmp->nm_sockreq.nr_cred
604 			 * credentials.
605 			 */
606 			KASSERT(nmp->nm_sockreq.nr_cred != NULL,
607 			    ("newnfs_request: NULL nr_cred"));
608 			crfree(authcred);
609 			authcred = crhold(nmp->nm_sockreq.nr_cred);
610 		}
611 		if (NFSHASINTEGRITY(nmp))
612 			secflavour = RPCSEC_GSS_KRB5I;
613 		else if (NFSHASPRIVACY(nmp))
614 			secflavour = RPCSEC_GSS_KRB5P;
615 		else
616 			secflavour = RPCSEC_GSS_KRB5;
617 		srv_principal = NFSMNT_SRVKRBNAME(nmp);
618 	} else if (nmp != NULL && !NFSHASKERB(nmp) &&
619 	    nd->nd_procnum != NFSPROC_NULL &&
620 	    (nd->nd_flag & ND_USEGSSNAME) != 0) {
621 		/*
622 		 * Use the uid that did the mount when the RPC is doing
623 		 * NFSv4 system operations, as indicated by the
624 		 * ND_USEGSSNAME flag, for the AUTH_SYS case.
625 		 * The credentials in nm_sockreq.nr_cred were used for the
626 		 * mount.
627 		 */
628 		KASSERT(nmp->nm_sockreq.nr_cred != NULL,
629 		    ("newnfs_request: NULL nr_cred"));
630 		crfree(authcred);
631 		authcred = crhold(nmp->nm_sockreq.nr_cred);
632 	}
633 
634 	if (nmp != NULL) {
635 		bzero(&nf, sizeof(struct nfs_feedback_arg));
636 		nf.nf_mount = nmp;
637 		nf.nf_td = td;
638 		nf.nf_lastmsg = NFSD_MONOSEC -
639 		    ((nmp->nm_tprintf_delay)-(nmp->nm_tprintf_initial_delay));
640 	}
641 
642 	if (nd->nd_procnum == NFSPROC_NULL)
643 		auth = authnone_create();
644 	else if (usegssname) {
645 		/*
646 		 * For this case, the authenticator is held in the
647 		 * nfssockreq structure, so don't release the reference count
648 		 * held on it. --> Don't AUTH_DESTROY() it in this function.
649 		 */
650 		if (nrp->nr_auth == NULL)
651 			nrp->nr_auth = nfs_getauth(nrp, secflavour,
652 			    clnt_principal, srv_principal, NULL, authcred);
653 		else
654 			rpc_gss_refresh_auth_call(nrp->nr_auth);
655 		auth = nrp->nr_auth;
656 	} else
657 		auth = nfs_getauth(nrp, secflavour, NULL,
658 		    srv_principal, NULL, authcred);
659 	crfree(authcred);
660 	if (auth == NULL) {
661 		m_freem(nd->nd_mreq);
662 		if (set_sigset)
663 			newnfs_restore_sigmask(td, &oldset);
664 		return (EACCES);
665 	}
666 	bzero(&ext, sizeof(ext));
667 	ext.rc_auth = auth;
668 	if (nmp != NULL) {
669 		ext.rc_feedback = nfs_feedback;
670 		ext.rc_feedback_arg = &nf;
671 	}
672 
673 	procnum = nd->nd_procnum;
674 	if ((nd->nd_flag & ND_NFSV4) &&
675 	    nd->nd_procnum != NFSPROC_NULL &&
676 	    nd->nd_procnum != NFSV4PROC_CBCOMPOUND)
677 		procnum = NFSV4PROC_COMPOUND;
678 
679 	if (nmp != NULL) {
680 		NFSINCRGLOBAL(nfsstatsv1.rpcrequests);
681 
682 		/* Map the procnum to the old NFSv2 one, as required. */
683 		if ((nd->nd_flag & ND_NFSV2) != 0) {
684 			if (nd->nd_procnum < NFS_V3NPROCS)
685 				procnum = nfsv2_procid[nd->nd_procnum];
686 			else
687 				procnum = NFSV2PROC_NOOP;
688 		}
689 
690 		/*
691 		 * Now only used for the R_DONTRECOVER case, but until that is
692 		 * supported within the krpc code, I need to keep a queue of
693 		 * outstanding RPCs for nfsv4 client requests.
694 		 */
695 		if ((nd->nd_flag & ND_NFSV4) && procnum == NFSV4PROC_COMPOUND)
696 			rep = malloc(sizeof(struct nfsreq),
697 			    M_NFSDREQ, M_WAITOK);
698 #ifdef KDTRACE_HOOKS
699 		if (dtrace_nfscl_nfs234_start_probe != NULL) {
700 			uint32_t probe_id;
701 			int probe_procnum;
702 
703 			if (nd->nd_flag & ND_NFSV4) {
704 				probe_id =
705 				    nfscl_nfs4_start_probes[nd->nd_procnum];
706 				probe_procnum = nd->nd_procnum;
707 			} else if (nd->nd_flag & ND_NFSV3) {
708 				probe_id = nfscl_nfs3_start_probes[procnum];
709 				probe_procnum = procnum;
710 			} else {
711 				probe_id =
712 				    nfscl_nfs2_start_probes[nd->nd_procnum];
713 				probe_procnum = procnum;
714 			}
715 			if (probe_id != 0)
716 				(dtrace_nfscl_nfs234_start_probe)
717 				    (probe_id, vp, nd->nd_mreq, cred,
718 				     probe_procnum);
719 		}
720 #endif
721 	}
722 	freeslot = -1;		/* Set to slot that needs to be free'd */
723 tryagain:
724 	slot = -1;		/* Slot that needs a sequence# increment. */
725 	/*
726 	 * This timeout specifies when a new socket should be created,
727 	 * along with new xid values. For UDP, this should be done
728 	 * infrequently, since retransmits of RPC requests should normally
729 	 * use the same xid.
730 	 */
731 	if (nmp == NULL) {
732 		timo.tv_usec = 0;
733 		if (clp == NULL)
734 			timo.tv_sec = NFSV4_UPCALLTIMEO;
735 		else
736 			timo.tv_sec = NFSV4_CALLBACKTIMEO;
737 	} else {
738 		if (nrp->nr_sotype != SOCK_DGRAM) {
739 			timo.tv_usec = 0;
740 			if ((nmp->nm_flag & NFSMNT_NFSV4))
741 				timo.tv_sec = INT_MAX;
742 			else
743 				timo.tv_sec = NFS_TCPTIMEO;
744 		} else {
745 			if (NFSHASSOFT(nmp)) {
746 				/*
747 				 * CLSET_RETRIES is set to 2, so this should be
748 				 * half of the total timeout required.
749 				 */
750 				timeo = nmp->nm_retry * nmp->nm_timeo / 2;
751 				if (timeo < 1)
752 					timeo = 1;
753 				timo.tv_sec = timeo / NFS_HZ;
754 				timo.tv_usec = (timeo % NFS_HZ) * 1000000 /
755 				    NFS_HZ;
756 			} else {
757 				/* For UDP hard mounts, use a large value. */
758 				timo.tv_sec = NFS_MAXTIMEO / NFS_HZ;
759 				timo.tv_usec = 0;
760 			}
761 		}
762 
763 		if (rep != NULL) {
764 			rep->r_flags = 0;
765 			rep->r_nmp = nmp;
766 			/*
767 			 * Chain request into list of outstanding requests.
768 			 */
769 			NFSLOCKREQ();
770 			TAILQ_INSERT_TAIL(&nfsd_reqq, rep, r_chain);
771 			NFSUNLOCKREQ();
772 		}
773 	}
774 
775 	nd->nd_mrep = NULL;
776 	if (clp != NULL && sep != NULL)
777 		stat = clnt_bck_call(nrp->nr_client, &ext, procnum,
778 		    nd->nd_mreq, &nd->nd_mrep, timo, sep->nfsess_xprt);
779 	else
780 		stat = CLNT_CALL_MBUF(nrp->nr_client, &ext, procnum,
781 		    nd->nd_mreq, &nd->nd_mrep, timo);
782 	NFSCL_DEBUG(2, "clnt call=%d\n", stat);
783 
784 	if (rep != NULL) {
785 		/*
786 		 * RPC done, unlink the request.
787 		 */
788 		NFSLOCKREQ();
789 		TAILQ_REMOVE(&nfsd_reqq, rep, r_chain);
790 		NFSUNLOCKREQ();
791 	}
792 
793 	/*
794 	 * If there was a successful reply and a tprintf msg.
795 	 * tprintf a response.
796 	 */
797 	if (stat == RPC_SUCCESS) {
798 		error = 0;
799 	} else if (stat == RPC_TIMEDOUT) {
800 		NFSINCRGLOBAL(nfsstatsv1.rpctimeouts);
801 		error = ETIMEDOUT;
802 	} else if (stat == RPC_VERSMISMATCH) {
803 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
804 		error = EOPNOTSUPP;
805 	} else if (stat == RPC_PROGVERSMISMATCH) {
806 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
807 		error = EPROTONOSUPPORT;
808 	} else if (stat == RPC_INTR) {
809 		error = EINTR;
810 	} else if (stat == RPC_CANTSEND || stat == RPC_CANTRECV ||
811 	     stat == RPC_SYSTEMERROR) {
812 		if ((nd->nd_flag & ND_NFSV41) != 0 && nmp != NULL &&
813 		    nd->nd_procnum != NFSPROC_NULL) {
814 			/*
815 			 * The nfsess_defunct field is protected by
816 			 * the NFSLOCKMNT()/nm_mtx lock and not the
817 			 * nfsess_mtx lock to simplify its handling,
818 			 * for the MDS session. This lock is also
819 			 * sufficient for nfsess_sessionid, since it
820 			 * never changes in the structure.
821 			 */
822 			NFSLOCKCLSTATE();
823 			NFSLOCKMNT(nmp);
824 			/* The session must be marked defunct. */
825 			if (dssep == NULL) {
826 				/*
827 				 * This is either an MDS proxy operation or
828 				 * a client mount with "soft,retrans=N" options.
829 				 * Mark the MDS session defunct and initiate
830 				 * recovery, as required.
831 				 */
832 				NFSCL_DEBUG(1, "Failed soft proxy RPC\n");
833 				sep = NFSMNT_MDSSESSION(nmp);
834 				if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
835 				    NFSX_V4SESSIONID) == 0) {
836 					/* Initiate recovery. */
837 					sep->nfsess_defunct = 1;
838 					NFSCL_DEBUG(1, "Marked defunct\n");
839 					if (nmp->nm_clp != NULL) {
840 						nmp->nm_clp->nfsc_flags |=
841 						    NFSCLFLAGS_RECOVER;
842 						wakeup(nmp->nm_clp);
843 					}
844 				}
845 			} else {
846 				/*
847 				 * This is a client side DS RPC. Just mark
848 				 * the session defunct.  A subsequent LayoutGet
849 				 * should get a new session.
850 				 */
851 				NFSCL_DEBUG(1, "Failed client DS RPC\n");
852 				if (bcmp(dssep->nfsess_sessionid,
853 				    nd->nd_sequence, NFSX_V4SESSIONID) == 0) {
854 					/* Mark it defunct. */
855 					dssep->nfsess_defunct = 1;
856 					NFSCL_DEBUG(1, "Marked defunct\n");
857 				}
858 			}
859 			NFSUNLOCKMNT(nmp);
860 			NFSUNLOCKCLSTATE();
861 		}
862 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
863 		error = ENXIO;
864 	} else {
865 		NFSINCRGLOBAL(nfsstatsv1.rpcinvalid);
866 		error = EACCES;
867 	}
868 	if (error) {
869 		m_freem(nd->nd_mreq);
870 		if (usegssname == 0)
871 			AUTH_DESTROY(auth);
872 		if (rep != NULL)
873 			free(rep, M_NFSDREQ);
874 		if (set_sigset)
875 			newnfs_restore_sigmask(td, &oldset);
876 		return (error);
877 	}
878 
879 	KASSERT(nd->nd_mrep != NULL, ("mrep shouldn't be NULL if no error\n"));
880 
881 	/*
882 	 * Search for any mbufs that are not a multiple of 4 bytes long
883 	 * or with m_data not longword aligned.
884 	 * These could cause pointer alignment problems, so copy them to
885 	 * well aligned mbufs.
886 	 */
887 	newnfs_realign(&nd->nd_mrep, M_WAITOK);
888 	nd->nd_md = nd->nd_mrep;
889 	nd->nd_dpos = NFSMTOD(nd->nd_md, caddr_t);
890 	nd->nd_repstat = 0;
891 	if (nd->nd_procnum != NFSPROC_NULL &&
892 	    nd->nd_procnum != NFSV4PROC_CBNULL) {
893 		/* If sep == NULL, set it to the default in nmp. */
894 		if (sep == NULL && nmp != NULL)
895 			sep = nfsmnt_mdssession(nmp);
896 		/*
897 		 * and now the actual NFS xdr.
898 		 */
899 		NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
900 		nd->nd_repstat = fxdr_unsigned(u_int32_t, *tl);
901 		if (nd->nd_repstat >= 10000)
902 			NFSCL_DEBUG(1, "proc=%d reps=%d\n", (int)nd->nd_procnum,
903 			    (int)nd->nd_repstat);
904 
905 		/*
906 		 * Get rid of the tag, return count and SEQUENCE result for
907 		 * NFSv4.
908 		 */
909 		if ((nd->nd_flag & ND_NFSV4) != 0) {
910 			NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
911 			i = fxdr_unsigned(int, *tl);
912 			error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
913 			if (error)
914 				goto nfsmout;
915 			NFSM_DISSECT(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
916 			opcnt = fxdr_unsigned(int, *tl++);
917 			i = fxdr_unsigned(int, *tl++);
918 			j = fxdr_unsigned(int, *tl);
919 			if (j >= 10000)
920 				NFSCL_DEBUG(1, "fop=%d fst=%d\n", i, j);
921 			/*
922 			 * If the first op is Sequence, free up the slot.
923 			 */
924 			if ((nmp != NULL && i == NFSV4OP_SEQUENCE && j != 0) ||
925 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE && j != 0))
926 				NFSCL_DEBUG(1, "failed seq=%d\n", j);
927 			if (((nmp != NULL && i == NFSV4OP_SEQUENCE && j == 0) ||
928 			    (clp != NULL && i == NFSV4OP_CBSEQUENCE &&
929 			    j == 0)) && sep != NULL) {
930 				if (i == NFSV4OP_SEQUENCE)
931 					NFSM_DISSECT(tl, uint32_t *,
932 					    NFSX_V4SESSIONID +
933 					    5 * NFSX_UNSIGNED);
934 				else
935 					NFSM_DISSECT(tl, uint32_t *,
936 					    NFSX_V4SESSIONID +
937 					    4 * NFSX_UNSIGNED);
938 				mtx_lock(&sep->nfsess_mtx);
939 				if (bcmp(tl, sep->nfsess_sessionid,
940 				    NFSX_V4SESSIONID) == 0) {
941 					tl += NFSX_V4SESSIONID / NFSX_UNSIGNED;
942 					retseq = fxdr_unsigned(uint32_t, *tl++);
943 					slot = fxdr_unsigned(int, *tl++);
944 					freeslot = slot;
945 					if (retseq != sep->nfsess_slotseq[slot])
946 						printf("retseq diff 0x%x\n",
947 						    retseq);
948 					retval = fxdr_unsigned(uint32_t, *++tl);
949 					if ((retval + 1) < sep->nfsess_foreslots
950 					    )
951 						sep->nfsess_foreslots = (retval
952 						    + 1);
953 					else if ((retval + 1) >
954 					    sep->nfsess_foreslots)
955 						sep->nfsess_foreslots = (retval
956 						    < 64) ? (retval + 1) : 64;
957 				}
958 				mtx_unlock(&sep->nfsess_mtx);
959 
960 				/* Grab the op and status for the next one. */
961 				if (opcnt > 1) {
962 					NFSM_DISSECT(tl, uint32_t *,
963 					    2 * NFSX_UNSIGNED);
964 					i = fxdr_unsigned(int, *tl++);
965 					j = fxdr_unsigned(int, *tl);
966 				}
967 			}
968 		}
969 		if (nd->nd_repstat != 0) {
970 			if (nd->nd_repstat == NFSERR_BADSESSION &&
971 			    nmp != NULL && dssep == NULL &&
972 			    (nd->nd_flag & ND_NFSV41) != 0) {
973 				/*
974 				 * If this is a client side MDS RPC, mark
975 				 * the MDS session defunct and initiate
976 				 * recovery, as required.
977 				 * The nfsess_defunct field is protected by
978 				 * the NFSLOCKMNT()/nm_mtx lock and not the
979 				 * nfsess_mtx lock to simplify its handling,
980 				 * for the MDS session. This lock is also
981 				 * sufficient for nfsess_sessionid, since it
982 				 * never changes in the structure.
983 				 */
984 				NFSCL_DEBUG(1, "Got badsession\n");
985 				NFSLOCKCLSTATE();
986 				NFSLOCKMNT(nmp);
987 				sep = NFSMNT_MDSSESSION(nmp);
988 				if (bcmp(sep->nfsess_sessionid, nd->nd_sequence,
989 				    NFSX_V4SESSIONID) == 0) {
990 					/* Initiate recovery. */
991 					sep->nfsess_defunct = 1;
992 					NFSCL_DEBUG(1, "Marked defunct\n");
993 					if (nmp->nm_clp != NULL) {
994 						nmp->nm_clp->nfsc_flags |=
995 						    NFSCLFLAGS_RECOVER;
996 						wakeup(nmp->nm_clp);
997 					}
998 				}
999 				NFSUNLOCKCLSTATE();
1000 				/*
1001 				 * Sleep for up to 1sec waiting for a new
1002 				 * session.
1003 				 */
1004 				mtx_sleep(&nmp->nm_sess, &nmp->nm_mtx, PZERO,
1005 				    "nfsbadsess", hz);
1006 				/*
1007 				 * Get the session again, in case a new one
1008 				 * has been created during the sleep.
1009 				 */
1010 				sep = NFSMNT_MDSSESSION(nmp);
1011 				NFSUNLOCKMNT(nmp);
1012 				if ((nd->nd_flag & ND_LOOPBADSESS) != 0) {
1013 					reterr = nfsv4_sequencelookup(nmp, sep,
1014 					    &slotpos, &maxslot, &slotseq,
1015 					    sessionid);
1016 					if (reterr == 0) {
1017 						/* Fill in new session info. */
1018 						NFSCL_DEBUG(1,
1019 						  "Filling in new sequence\n");
1020 						tl = nd->nd_sequence;
1021 						bcopy(sessionid, tl,
1022 						    NFSX_V4SESSIONID);
1023 						tl += NFSX_V4SESSIONID /
1024 						    NFSX_UNSIGNED;
1025 						*tl++ = txdr_unsigned(slotseq);
1026 						*tl++ = txdr_unsigned(slotpos);
1027 						*tl = txdr_unsigned(maxslot);
1028 					}
1029 					if (reterr == NFSERR_BADSESSION ||
1030 					    reterr == 0) {
1031 						NFSCL_DEBUG(1,
1032 						    "Badsession looping\n");
1033 						m_freem(nd->nd_mrep);
1034 						nd->nd_mrep = NULL;
1035 						goto tryagain;
1036 					}
1037 					nd->nd_repstat = reterr;
1038 					NFSCL_DEBUG(1, "Got err=%d\n", reterr);
1039 				}
1040 			}
1041 			if (((nd->nd_repstat == NFSERR_DELAY ||
1042 			      nd->nd_repstat == NFSERR_GRACE) &&
1043 			     (nd->nd_flag & ND_NFSV4) &&
1044 			     nd->nd_procnum != NFSPROC_DELEGRETURN &&
1045 			     nd->nd_procnum != NFSPROC_SETATTR &&
1046 			     nd->nd_procnum != NFSPROC_READ &&
1047 			     nd->nd_procnum != NFSPROC_READDS &&
1048 			     nd->nd_procnum != NFSPROC_WRITE &&
1049 			     nd->nd_procnum != NFSPROC_WRITEDS &&
1050 			     nd->nd_procnum != NFSPROC_OPEN &&
1051 			     nd->nd_procnum != NFSPROC_CREATE &&
1052 			     nd->nd_procnum != NFSPROC_OPENCONFIRM &&
1053 			     nd->nd_procnum != NFSPROC_OPENDOWNGRADE &&
1054 			     nd->nd_procnum != NFSPROC_CLOSE &&
1055 			     nd->nd_procnum != NFSPROC_LOCK &&
1056 			     nd->nd_procnum != NFSPROC_LOCKU) ||
1057 			    (nd->nd_repstat == NFSERR_DELAY &&
1058 			     (nd->nd_flag & ND_NFSV4) == 0) ||
1059 			    nd->nd_repstat == NFSERR_RESOURCE) {
1060 				if (trylater_delay > NFS_TRYLATERDEL)
1061 					trylater_delay = NFS_TRYLATERDEL;
1062 				waituntil = NFSD_MONOSEC + trylater_delay;
1063 				while (NFSD_MONOSEC < waituntil)
1064 					(void) nfs_catnap(PZERO, 0, "nfstry");
1065 				trylater_delay *= 2;
1066 				if (slot != -1) {
1067 					mtx_lock(&sep->nfsess_mtx);
1068 					sep->nfsess_slotseq[slot]++;
1069 					*nd->nd_slotseq = txdr_unsigned(
1070 					    sep->nfsess_slotseq[slot]);
1071 					mtx_unlock(&sep->nfsess_mtx);
1072 				}
1073 				m_freem(nd->nd_mrep);
1074 				nd->nd_mrep = NULL;
1075 				goto tryagain;
1076 			}
1077 
1078 			/*
1079 			 * If the File Handle was stale, invalidate the
1080 			 * lookup cache, just in case.
1081 			 * (vp != NULL implies a client side call)
1082 			 */
1083 			if (nd->nd_repstat == ESTALE && vp != NULL) {
1084 				cache_purge(vp);
1085 				if (ncl_call_invalcaches != NULL)
1086 					(*ncl_call_invalcaches)(vp);
1087 			}
1088 		}
1089 		if ((nd->nd_flag & ND_NFSV4) != 0) {
1090 			/* Free the slot, as required. */
1091 			if (freeslot != -1)
1092 				nfsv4_freeslot(sep, freeslot);
1093 			/*
1094 			 * If this op is Putfh, throw its results away.
1095 			 */
1096 			if (j >= 10000)
1097 				NFSCL_DEBUG(1, "nop=%d nst=%d\n", i, j);
1098 			if (nmp != NULL && i == NFSV4OP_PUTFH && j == 0) {
1099 				NFSM_DISSECT(tl,u_int32_t *,2 * NFSX_UNSIGNED);
1100 				i = fxdr_unsigned(int, *tl++);
1101 				j = fxdr_unsigned(int, *tl);
1102 				if (j >= 10000)
1103 					NFSCL_DEBUG(1, "n2op=%d n2st=%d\n", i,
1104 					    j);
1105 				/*
1106 				 * All Compounds that do an Op that must
1107 				 * be in sequence consist of NFSV4OP_PUTFH
1108 				 * followed by one of these. As such, we
1109 				 * can determine if the seqid# should be
1110 				 * incremented, here.
1111 				 */
1112 				if ((i == NFSV4OP_OPEN ||
1113 				     i == NFSV4OP_OPENCONFIRM ||
1114 				     i == NFSV4OP_OPENDOWNGRADE ||
1115 				     i == NFSV4OP_CLOSE ||
1116 				     i == NFSV4OP_LOCK ||
1117 				     i == NFSV4OP_LOCKU) &&
1118 				    (j == 0 ||
1119 				     (j != NFSERR_STALECLIENTID &&
1120 				      j != NFSERR_STALESTATEID &&
1121 				      j != NFSERR_BADSTATEID &&
1122 				      j != NFSERR_BADSEQID &&
1123 				      j != NFSERR_BADXDR &&
1124 				      j != NFSERR_RESOURCE &&
1125 				      j != NFSERR_NOFILEHANDLE)))
1126 					nd->nd_flag |= ND_INCRSEQID;
1127 			}
1128 			/*
1129 			 * If this op's status is non-zero, mark
1130 			 * that there is no more data to process.
1131 			 * The exception is Setattr, which always has xdr
1132 			 * when it has failed.
1133 			 */
1134 			if (j != 0 && i != NFSV4OP_SETATTR)
1135 				nd->nd_flag |= ND_NOMOREDATA;
1136 
1137 			/*
1138 			 * If R_DONTRECOVER is set, replace the stale error
1139 			 * reply, so that recovery isn't initiated.
1140 			 */
1141 			if ((nd->nd_repstat == NFSERR_STALECLIENTID ||
1142 			     nd->nd_repstat == NFSERR_BADSESSION ||
1143 			     nd->nd_repstat == NFSERR_STALESTATEID) &&
1144 			    rep != NULL && (rep->r_flags & R_DONTRECOVER))
1145 				nd->nd_repstat = NFSERR_STALEDONTRECOVER;
1146 		}
1147 	}
1148 
1149 #ifdef KDTRACE_HOOKS
1150 	if (nmp != NULL && dtrace_nfscl_nfs234_done_probe != NULL) {
1151 		uint32_t probe_id;
1152 		int probe_procnum;
1153 
1154 		if (nd->nd_flag & ND_NFSV4) {
1155 			probe_id = nfscl_nfs4_done_probes[nd->nd_procnum];
1156 			probe_procnum = nd->nd_procnum;
1157 		} else if (nd->nd_flag & ND_NFSV3) {
1158 			probe_id = nfscl_nfs3_done_probes[procnum];
1159 			probe_procnum = procnum;
1160 		} else {
1161 			probe_id = nfscl_nfs2_done_probes[nd->nd_procnum];
1162 			probe_procnum = procnum;
1163 		}
1164 		if (probe_id != 0)
1165 			(dtrace_nfscl_nfs234_done_probe)(probe_id, vp,
1166 			    nd->nd_mreq, cred, probe_procnum, 0);
1167 	}
1168 #endif
1169 
1170 	m_freem(nd->nd_mreq);
1171 	if (usegssname == 0)
1172 		AUTH_DESTROY(auth);
1173 	if (rep != NULL)
1174 		free(rep, M_NFSDREQ);
1175 	if (set_sigset)
1176 		newnfs_restore_sigmask(td, &oldset);
1177 	return (0);
1178 nfsmout:
1179 	mbuf_freem(nd->nd_mrep);
1180 	mbuf_freem(nd->nd_mreq);
1181 	if (usegssname == 0)
1182 		AUTH_DESTROY(auth);
1183 	if (rep != NULL)
1184 		free(rep, M_NFSDREQ);
1185 	if (set_sigset)
1186 		newnfs_restore_sigmask(td, &oldset);
1187 	return (error);
1188 }
1189 
1190 /*
1191  * Mark all of an nfs mount's outstanding requests with R_SOFTTERM and
1192  * wait for all requests to complete. This is used by forced unmounts
1193  * to terminate any outstanding RPCs.
1194  */
1195 int
1196 newnfs_nmcancelreqs(struct nfsmount *nmp)
1197 {
1198 	struct nfsclds *dsp;
1199 	struct __rpc_client *cl;
1200 
1201 	if (nmp->nm_sockreq.nr_client != NULL)
1202 		CLNT_CLOSE(nmp->nm_sockreq.nr_client);
1203 lookformore:
1204 	NFSLOCKMNT(nmp);
1205 	TAILQ_FOREACH(dsp, &nmp->nm_sess, nfsclds_list) {
1206 		NFSLOCKDS(dsp);
1207 		if (dsp != TAILQ_FIRST(&nmp->nm_sess) &&
1208 		    (dsp->nfsclds_flags & NFSCLDS_CLOSED) == 0 &&
1209 		    dsp->nfsclds_sockp != NULL &&
1210 		    dsp->nfsclds_sockp->nr_client != NULL) {
1211 			dsp->nfsclds_flags |= NFSCLDS_CLOSED;
1212 			cl = dsp->nfsclds_sockp->nr_client;
1213 			NFSUNLOCKDS(dsp);
1214 			NFSUNLOCKMNT(nmp);
1215 			CLNT_CLOSE(cl);
1216 			goto lookformore;
1217 		}
1218 		NFSUNLOCKDS(dsp);
1219 	}
1220 	NFSUNLOCKMNT(nmp);
1221 	return (0);
1222 }
1223 
1224 /*
1225  * Any signal that can interrupt an NFS operation in an intr mount
1226  * should be added to this set. SIGSTOP and SIGKILL cannot be masked.
1227  */
1228 int newnfs_sig_set[] = {
1229 	SIGINT,
1230 	SIGTERM,
1231 	SIGHUP,
1232 	SIGKILL,
1233 	SIGQUIT
1234 };
1235 
1236 /*
1237  * Check to see if one of the signals in our subset is pending on
1238  * the process (in an intr mount).
1239  */
1240 static int
1241 nfs_sig_pending(sigset_t set)
1242 {
1243 	int i;
1244 
1245 	for (i = 0 ; i < nitems(newnfs_sig_set); i++)
1246 		if (SIGISMEMBER(set, newnfs_sig_set[i]))
1247 			return (1);
1248 	return (0);
1249 }
1250 
1251 /*
1252  * The set/restore sigmask functions are used to (temporarily) overwrite
1253  * the thread td_sigmask during an RPC call (for example). These are also
1254  * used in other places in the NFS client that might tsleep().
1255  */
1256 void
1257 newnfs_set_sigmask(struct thread *td, sigset_t *oldset)
1258 {
1259 	sigset_t newset;
1260 	int i;
1261 	struct proc *p;
1262 
1263 	SIGFILLSET(newset);
1264 	if (td == NULL)
1265 		td = curthread; /* XXX */
1266 	p = td->td_proc;
1267 	/* Remove the NFS set of signals from newset */
1268 	PROC_LOCK(p);
1269 	mtx_lock(&p->p_sigacts->ps_mtx);
1270 	for (i = 0 ; i < nitems(newnfs_sig_set); i++) {
1271 		/*
1272 		 * But make sure we leave the ones already masked
1273 		 * by the process, ie. remove the signal from the
1274 		 * temporary signalmask only if it wasn't already
1275 		 * in p_sigmask.
1276 		 */
1277 		if (!SIGISMEMBER(td->td_sigmask, newnfs_sig_set[i]) &&
1278 		    !SIGISMEMBER(p->p_sigacts->ps_sigignore, newnfs_sig_set[i]))
1279 			SIGDELSET(newset, newnfs_sig_set[i]);
1280 	}
1281 	mtx_unlock(&p->p_sigacts->ps_mtx);
1282 	kern_sigprocmask(td, SIG_SETMASK, &newset, oldset,
1283 	    SIGPROCMASK_PROC_LOCKED);
1284 	PROC_UNLOCK(p);
1285 }
1286 
1287 void
1288 newnfs_restore_sigmask(struct thread *td, sigset_t *set)
1289 {
1290 	if (td == NULL)
1291 		td = curthread; /* XXX */
1292 	kern_sigprocmask(td, SIG_SETMASK, set, NULL, 0);
1293 }
1294 
1295 /*
1296  * NFS wrapper to msleep(), that shoves a new p_sigmask and restores the
1297  * old one after msleep() returns.
1298  */
1299 int
1300 newnfs_msleep(struct thread *td, void *ident, struct mtx *mtx, int priority, char *wmesg, int timo)
1301 {
1302 	sigset_t oldset;
1303 	int error;
1304 
1305 	if ((priority & PCATCH) == 0)
1306 		return msleep(ident, mtx, priority, wmesg, timo);
1307 	if (td == NULL)
1308 		td = curthread; /* XXX */
1309 	newnfs_set_sigmask(td, &oldset);
1310 	error = msleep(ident, mtx, priority, wmesg, timo);
1311 	newnfs_restore_sigmask(td, &oldset);
1312 	return (error);
1313 }
1314 
1315 /*
1316  * Test for a termination condition pending on the process.
1317  * This is used for NFSMNT_INT mounts.
1318  */
1319 int
1320 newnfs_sigintr(struct nfsmount *nmp, struct thread *td)
1321 {
1322 	struct proc *p;
1323 	sigset_t tmpset;
1324 
1325 	/* Terminate all requests while attempting a forced unmount. */
1326 	if (NFSCL_FORCEDISM(nmp->nm_mountp))
1327 		return (EIO);
1328 	if (!(nmp->nm_flag & NFSMNT_INT))
1329 		return (0);
1330 	if (td == NULL)
1331 		return (0);
1332 	p = td->td_proc;
1333 	PROC_LOCK(p);
1334 	tmpset = p->p_siglist;
1335 	SIGSETOR(tmpset, td->td_siglist);
1336 	SIGSETNAND(tmpset, td->td_sigmask);
1337 	mtx_lock(&p->p_sigacts->ps_mtx);
1338 	SIGSETNAND(tmpset, p->p_sigacts->ps_sigignore);
1339 	mtx_unlock(&p->p_sigacts->ps_mtx);
1340 	if ((SIGNOTEMPTY(p->p_siglist) || SIGNOTEMPTY(td->td_siglist))
1341 	    && nfs_sig_pending(tmpset)) {
1342 		PROC_UNLOCK(p);
1343 		return (EINTR);
1344 	}
1345 	PROC_UNLOCK(p);
1346 	return (0);
1347 }
1348 
1349 static int
1350 nfs_msg(struct thread *td, const char *server, const char *msg, int error)
1351 {
1352 	struct proc *p;
1353 
1354 	p = td ? td->td_proc : NULL;
1355 	if (error) {
1356 		tprintf(p, LOG_INFO, "nfs server %s: %s, error %d\n",
1357 		    server, msg, error);
1358 	} else {
1359 		tprintf(p, LOG_INFO, "nfs server %s: %s\n", server, msg);
1360 	}
1361 	return (0);
1362 }
1363 
1364 static void
1365 nfs_down(struct nfsmount *nmp, struct thread *td, const char *msg,
1366     int error, int flags)
1367 {
1368 	if (nmp == NULL)
1369 		return;
1370 	mtx_lock(&nmp->nm_mtx);
1371 	if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) {
1372 		nmp->nm_state |= NFSSTA_TIMEO;
1373 		mtx_unlock(&nmp->nm_mtx);
1374 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1375 		    VQ_NOTRESP, 0);
1376 	} else
1377 		mtx_unlock(&nmp->nm_mtx);
1378 	mtx_lock(&nmp->nm_mtx);
1379 	if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1380 		nmp->nm_state |= NFSSTA_LOCKTIMEO;
1381 		mtx_unlock(&nmp->nm_mtx);
1382 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1383 		    VQ_NOTRESPLOCK, 0);
1384 	} else
1385 		mtx_unlock(&nmp->nm_mtx);
1386 	nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, error);
1387 }
1388 
1389 static void
1390 nfs_up(struct nfsmount *nmp, struct thread *td, const char *msg,
1391     int flags, int tprintfmsg)
1392 {
1393 	if (nmp == NULL)
1394 		return;
1395 	if (tprintfmsg) {
1396 		nfs_msg(td, nmp->nm_mountp->mnt_stat.f_mntfromname, msg, 0);
1397 	}
1398 
1399 	mtx_lock(&nmp->nm_mtx);
1400 	if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) {
1401 		nmp->nm_state &= ~NFSSTA_TIMEO;
1402 		mtx_unlock(&nmp->nm_mtx);
1403 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1404 		    VQ_NOTRESP, 1);
1405 	} else
1406 		mtx_unlock(&nmp->nm_mtx);
1407 
1408 	mtx_lock(&nmp->nm_mtx);
1409 	if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) {
1410 		nmp->nm_state &= ~NFSSTA_LOCKTIMEO;
1411 		mtx_unlock(&nmp->nm_mtx);
1412 		vfs_event_signal(&nmp->nm_mountp->mnt_stat.f_fsid,
1413 		    VQ_NOTRESPLOCK, 1);
1414 	} else
1415 		mtx_unlock(&nmp->nm_mtx);
1416 }
1417 
1418